Data Aware Neural Architecture Search for Resource Constrained Systems

Watch a 14-minute research symposium presentation exploring Data Aware Neural Architecture Search (NAS) for resource-constrained machine learning systems. Learn how PhD student Emil NJOR from Technical University of Denmark proposes enhancing traditional NAS approaches by incorporating input data granularity considerations. Discover how this novel "Data Aware NAS" method compares to conventional NAS techniques, particularly when optimizing neural network architectures for multiple metrics like accuracy and memory constraints in embedded systems. Understand why single-metric optimization is insufficient for resource-limited devices and how considering data characteristics can lead to more efficient architecture search outcomes.